Abstract

Reinforcement Learning with A* and a Deep Heuristic

Inspired by the recent achievements of methods which combine trees and DNNs (e.g. AlphaZero), this study demonstrates how to effectively combine AI with a deep heuristic represented by a DNN. AI is a highly popular and successful path-planning algorithm, but it's usability is limited to only those domains where a good heuristic is known. This new algorithm, which we call Aleph-0, enables us to replicate the success of AI in new domains, including that of autonomous vehicles.


Session No: SIL8124
Speaker: Ariel Keselman
Type:

Autonomous Vehicle Technologies

Date: Thursday - October 18, 2018 05:00 PM - 05:45 PM
Location: Hall B
Topics: Deep Learning and AI, Autonomous Vehicle Technologies
Industry: Automotive / Transportation

Inspired by the recent achievements of methods which combine trees and DNNs (e.g. AlphaZero), this study demonstrates how to effectively combine AI with a deep heuristic represented by a DNN. AI is a highly popular and successful path-planning algorithm, but it's usability is limited to only those domains where a good heuristic is known. This new algorithm, which we call Aleph-0, enables us to replicate the success of AI in new domains, including that of autonomous vehicles.